Transformers - The Batch | DeepLearning.AI

Graph depicts Qwen3.8-Max's rising score from 0.474 to peak at 0.725 during RL training environments.

Machine Learning Research

Qwen3.8-Max Lands With A Bang: Inside Alibaba's two new models, Qwen3.8-Max and Qwen3.8-27B

Open models are getting larger and more capable. A few weeks ago, Moonshot AI announced Kimi K3, the largest and best-performing open weights model yet. Last week, Alibaba answered by releasing weights for a giant of its own.

Diagram shows Minimax H3's process from text prompt input to audiovisual output via a transformer.

Machine Learning Research

Minimax’s State-of-the-Art Video Model Is Only Minimally Open: Minimax H3's weights are free, but carry unusual restrictions

A free-to-download model sets a new standard for video generation and editing, but its license comes with unexpected restrictions.

A chart shows DeepSeek-V4-Flash leading in a green area with low cost and strong performance, marked in blue.

Machine Learning Research

DeepSeek Pushes the Frontier Again: DeepSeek refreshed its V4 Flash model with an impressive fine-tune

DeepSeek’s updated small model overtook the company’s own flagship.

Kimi diagram of open frontier model size over time, with Moonshot AI's Kimi K3 at 2.8 trillion parameters, ahead of DeepSeek at 1.6T parameters

Machine Learning Research

Kimi K3 Reveals How A Giant Frontier AI Model Works: Moonshot's latest model outshines all but GPT-5.6 Sol and Claude 5 Fable

Moonshot’s latest model leapfrogged the month-old GLM-5.2 and a host of proprietary competitors to finish just behind GPT-5.6 Sol and Claude Fable 5 on many benchmarks.

Flowchart of an ESMC-6B model with sequence encoding layers, language model, and diffusion transformer output.

Machine Learning Research

Biological Molecules as Language: ESMFold2 approaches AlphaFold 3 performance but with an open, Transformer-based architecture

Google’s AlphaFold models pioneered the task of finding the shapes of biologically active molecules, opening new pathways for drug development.

AFM 3 Core model architecture visualizes DRAM and NAND processes in AI with focus on sparsely-activated LLM operations.

Machine Learning Research

Large-Model AI for Apple Devices: 2026's Apple Foundation Models bring AI to MacBooks, iPhones, and the cloud

The third generation of Apple Foundation Models — fruit of Apple’s collaboration with Google — introduces a variation on the mixture-of-experts architecture that runs on local devices.

Performance data table displays metrics for conversational models, emphasizing TML-Interaction-Small's results.

Machine Learning Research

Built-In Conversational Interactivity: Thinking Machines reveals its first interaction model, a new type of multimodal AI

Conversational models typically wait for a turn before they respond.

GIF showing fluid simulation with patch jittering, followed by predicted vs. actual flows across time steps 0–38.

Machine Learning Research

How Liquids and Gases Behave: A dynamic fluids model appears to solve transformers’ pixellation problem

Simulating complex physical systems through traditional numerical methods is slow and expensive, and simulations based on machine learning are usually specialized for a specific type of system, such as water in a pipe or atmosphere surrounding a planet.

Diagram shows DNA analysis with interconnected devices, output types, and species-specific data.

Science

Dark DNA Unveiled: Google’s AlphaGenome interprets DNA that regulates genetic expression

An open-weights model could help scientists compare the impact of genetic variations, identify mutations that cause diseases, and develop treatments.

Two graphs show TTT-E2E maintains stable loss and latency across increasing context lengths up to 128k.

Machine Learning Research

Learning Long Context at Inference: Test-Time Training End-to-End (TTT-E2E) retrains model weights to handle long inputs

Large language models typically become less accurate and slower when they process longer contexts, but researchers enabled an LLM to keep accuracy stable and inference time constant as its context grew.

Machine Learning Research

A Single Tokenizer for Visual Media: Apple’s AToken, a multimodal model with a single encoder and tokenizer for images, videos, and 3D objects

Multimodal models typically use different tokenizers to embed different media types, and different encoders when training to generate media rather than classify it.

Diagram shows SleepFM's data processing flow from sleep signals to disease prediction using neural networks.

Machine Learning Research

Sleep Signals Predict Illness: SleepFM detects signs of neurological disorders years before symptoms manifest

Difficulty sleeping often precedes heart disease, psychiatric disorders, and many other illnesses. Researchers used data gathered during sleep studies to detect such conditions.